PREDICTIVE MAINTENANCE
Projects aimed at improving the country's energy supply poses several challenges for the energy industry. The most important of them include increasing the efficiency of existing power units, their modernisation, extending the service life of ageing power plants, putting into operation equipment and building new-generation power units.

The experience of operating TPPs and NPPs shows that most cases of plant downtime are associated with the failure of thermal power equipment. Moreover, this applies not only to leading equipment (steam generators, turbines), but also auxiliary equipment and electronics.
Python / Machine Learning / IOT / Mathematical Modelling / TensorFlow / AWS / Azure / Docker / CI/CD
MACHINES LIVE LONGER
THE CHALLENGE
The objective of this project was to develop a system for predicting the failure of various equipment. In our case, the subject of forecasting is the equipment state at thermal power plants. Informing of employees at the right time makes it possible in time to identify problem areas of the system and take the necessary measures.

The goals set by the customer were the following:

  • inform about the need to check specific equipment one hour before a possible breakdown;
  • reduce the time spent on diagnosis;
  • the economic effect of reducing the number of fines (non-compliance with the plan);
  • reduction of equipment downtime due to breakdowns;
THE SOLUTION
THE RESULT
Due to the absence of a labelled dataset with breakdowns, we decided to build alternative algorithms for finding faults without labelling. Therefore, the solution to this problem was based on the anomaly detector. It monitored more than 2000 signals (temperature, pressure, water/gas flow rate, etc.) from sensors at TPPs. The signals were updated with a minutely frequency.

The anomaly detection model is based on a convolutional neural network (autoencoder). As the output of the model we get the localisation of possible breakdowns with associated probability predicted for 1 hour ahead. Based on this information, the algorithm makes a recommendation to the experts. All data was visualised in BI about the operation of equipment for convenient monitoring by the management and employees of the thermal power plants.
After the implementation of the system, the client received:

  • Reduced equipment downtime due to breakdowns by 20%;
  • Service time decreased by 30%;
  • Reduced time to detect breakdowns hundreds of times :)
  • We also configured the system for collecting and storing data on the operation of the equipment.
REQUEST
AI AT WORK
MEASURED
Ukraien, Kiev
Cupertino str. 132
info@neurons-lab.com
+44 330 027 2146
International House,
64 Nile Str, London, N1 7SR,
United Kingdom
© 2020 Neurons Lab LTD. Registered in England and Wales
Company Number 12265479